Add official FunASR runtime websocket service

README.md

@@ -43,7 +43,8 @@ docker compose down
 | --- | --- | --- | --- |
 | Speakr | `voice-speakr` | `8899` | 页面和后端 |
 | ASR | `asr-test` | `59805` | Paraformer HTTP / WebSocket 识别 |
-| 官方实时 ASR | `funasr-online` | `10096` | FunASR 官方实时 WebSocket |
+| Python 实时 ASR | `funasr-online` | `10096` | FunASR Python WebSocket，保留用于回退 |
+| 官方 Runtime 实时 ASR | `funasr-runtime-2pass` | `10097` | FunASR 官方 C++/ONNX 2pass WebSocket |
 | 声纹提取 | `voice-dected` | `18000` | `/extract_embedding` 声纹向量服务 |
 | 实时桥接 | `fastapi-wss` | `10095` | 页面实时 ASR WebSocket 桥接 |
 | Redis | `voice-redis` | `16380` | 实时桥接的缓存/会话存储 |
@@ -86,7 +87,9 @@ docker compose up -d
 ```env
 ASR_BASE_URL=http://asr-test:59805
 FUNASR_WEBSOCKET_IP=ws://fastapi-wss:10095/websocket_offline
-TARGET_WS_URL=ws://funasr-online:10096
+TARGET_WS_URL=ws://funasr-runtime-2pass:10095
+# 如需回退到 Python 版实时 ASR，可改为：
+# TARGET_WS_URL=ws://funasr-online:10096
 VOICE_DETECTED_URL=http://voice-dected:8000
 SPEAKER_DET_URL=http://voice-dected:8000/extract_embedding
 REDIS_URL=redis://redis:6379/0

config.env

@@ -14,8 +14,10 @@ TEXT_MODEL_API_KEY=none
 
 # Realtime WebSocket bridge
 FUNASR_WEBSOCKET_IP=ws://fastapi-wss:10095/websocket_offline
-TARGET_WS_URL=ws://funasr-online:10096
+TARGET_WS_URL=ws://funasr-runtime-2pass:10095
-REDIS_URL=redis://redis:6379/0
+# Optional A/B target for the official C++/ONNX FunASR runtime service:
+# TARGET_WS_URL=ws://funasr-online:10096
+REDIS_URL=redis://redis:6379/0
 LLM_API_KEY=none
 LLM_BASE_URL=http://192.168.0.46:59800/v1
 LLM_MODEL=46-qwen3.5-35B

docker-compose.yml

@@ -94,9 +94,9 @@ services:
       - "1"
 
 
-  funasr-online:
+  funasr-online:
-    image: paraformer-asr:latest
+    image: paraformer-asr:latest
-    container_name: funasr-online
+    container_name: funasr-online
     restart: unless-stopped
     runtime: nvidia
     gpus:
@@ -153,23 +153,53 @@ services:
       - --keyfile
       - ""
 
+  funasr-runtime-2pass:
+    image: registry.cn-hangzhou.aliyuncs.com/funasr_repo/funasr:funasr-runtime-sdk-online-cpu-0.1.13
+    container_name: funasr-runtime-2pass
+    restart: unless-stopped
+    working_dir: /workspace/FunASR/runtime
+    volumes:
+      - ./deploy/data/funasr-runtime-models:/workspace/models
+    ports:
+      - "10097:10095"
+    command:
+      - bash
+      - -lc
+      - >
+        touch /workspace/models/hotwords.txt &&
+        exec /workspace/FunASR/runtime/websocket/build/bin/funasr-wss-server-2pass
+        --download-model-dir /workspace/models
+        --vad-dir damo/speech_fsmn_vad_zh-cn-16k-common-onnx
+        --model-dir damo/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-onnx
+        --online-model-dir damo/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-online-onnx
+        --punc-dir damo/punc_ct-transformer_zh-cn-common-vad_realtime-vocab272727-onnx
+        --lm-dir damo/speech_ngram_lm_zh-cn-ai-wesp-fst
+        --itn-dir thuduj12/fst_itn_zh
+        --decoder-thread-num 4
+        --model-thread-num 1
+        --io-thread-num 1
+        --port 10095
+        --hotword /workspace/models/hotwords.txt
+        --certfile ""
+        --keyfile ""
+
   fastapi-wss:
-    image: fastapi-wss:runtime
+    image: fastapi-wss:runtime
-    container_name: fastapi-wss
+    container_name: fastapi-wss
     restart: unless-stopped
     working_dir: /app/src
-    env_file:
+    env_file:
-      - ./config.env
+      - ./config.env
-    environment:
+    environment:
-      TARGET_WS_URL: ws://funasr-online:10096
+      TARGET_WS_URL: ws://funasr-runtime-2pass:10095
-      REDIS_URL: redis://redis:6379/0
+      REDIS_URL: redis://redis:6379/0
     volumes:
       - ./fastapi_wss:/app
     ports:
       - "10095:10095"
-    depends_on:
+    depends_on:
-      - funasr-online
+      - funasr-runtime-2pass
-      - redis
+      - redis
     command:
       - python
       - a2a_wss.py

fastapi_wss/src/a2a_wss.py

@@ -19,6 +19,7 @@ import os
 import websockets
 import re
 import logging
+from contextlib import asynccontextmanager
 from pathlib import Path
 from typing import List, Optional, Dict, Set
 from dataclasses import dataclass
@@ -789,6 +790,33 @@ def build_role_enriched_message(speaker_msg: Dict, role_result: Dict) -> Dict:
 # 5. WebSocket 处理与任务追踪
 # ========================
 
+@asynccontextmanager
+async def connect_target_asr():
+    connect_options = {
+        "close_timeout": 10,
+        "ping_interval": None,
+        "ping_timeout": None,
+        "max_queue": 1024,
+    }
+    try:
+        target_ws = await websockets.connect(
+            TARGET_WS_URL,
+            **connect_options,
+            subprotocols=["binary"],
+        )
+    except websockets.exceptions.InvalidMessage as exc:
+        logger.warning(
+            "ASR WebSocket binary subprotocol handshake failed, retrying without subprotocol: %s",
+            exc,
+        )
+        target_ws = await websockets.connect(TARGET_WS_URL, **connect_options)
+
+    try:
+        yield target_ws
+    finally:
+        await target_ws.close()
+
+
 async def forward_audio(websocket):
     """
     WebSocket 连接处理函数（核心入口）
@@ -824,14 +852,7 @@ async def forward_audio(websocket):
     try:
         # 连接到目标 ASR 服务
         # 添加连接参数以提高鲁棒性
-        async with websockets.connect(
+        async with connect_target_asr() as target_ws:
-            TARGET_WS_URL,
-            close_timeout=10,  # 关闭超时
-            ping_interval=None,
-            ping_timeout=None,
-            max_queue=1024,     # 消息队列大小
-            subprotocols=["binary"],
-        ) as target_ws:
             logger.info(f"已连接到 ASR 服务: {TARGET_WS_URL}")
 
             # 发送 ASR 配置（双通道模式，PCM 音频格式）